1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
|
// Copyright 2018 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
namespace VideoCommon::Shader {
using Tegra::Shader::ConditionCode;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
using Tegra::Shader::Register;
using Tegra::Shader::SystemVariable;
u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
switch (opcode->get().GetId()) {
case OpCode::Id::NOP: {
UNIMPLEMENTED_IF(instr.nop.cc != Tegra::Shader::ConditionCode::T);
UNIMPLEMENTED_IF(instr.nop.trigger != 0);
// With the previous preconditions, this instruction is a no-operation.
break;
}
case OpCode::Id::EXIT: {
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "EXIT condition code used: {}",
static_cast<u32>(cc));
switch (instr.flow.cond) {
case Tegra::Shader::FlowCondition::Always:
bb.push_back(Operation(OperationCode::Exit));
if (instr.pred.pred_index == static_cast<u64>(Tegra::Shader::Pred::UnusedIndex)) {
// If this is an unconditional exit then just end processing here,
// otherwise we have to account for the possibility of the condition
// not being met, so continue processing the next instruction.
pc = MAX_PROGRAM_LENGTH - 1;
}
break;
case Tegra::Shader::FlowCondition::Fcsm_Tr:
// TODO(bunnei): What is this used for? If we assume this conditon is not
// satisifed, dual vertex shaders in Farming Simulator make more sense
UNIMPLEMENTED_MSG("Skipping unknown FlowCondition::Fcsm_Tr");
break;
default:
UNIMPLEMENTED_MSG("Unhandled flow condition: {}",
static_cast<u32>(instr.flow.cond.Value()));
}
break;
}
case OpCode::Id::KIL: {
UNIMPLEMENTED_IF(instr.flow.cond != Tegra::Shader::FlowCondition::Always);
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "KIL condition code used: {}",
static_cast<u32>(cc));
bb.push_back(Operation(OperationCode::Discard));
break;
}
case OpCode::Id::MOV_SYS: {
const Node value = [this, instr] {
switch (instr.sys20) {
case SystemVariable::LaneId:
LOG_WARNING(HW_GPU, "MOV_SYS instruction with LaneId is incomplete");
return Immediate(0U);
case SystemVariable::InvocationId:
return Operation(OperationCode::InvocationId);
case SystemVariable::Ydirection:
return Operation(OperationCode::YNegate);
case SystemVariable::InvocationInfo:
LOG_WARNING(HW_GPU, "MOV_SYS instruction with InvocationInfo is incomplete");
return Immediate(0U);
case SystemVariable::Tid: {
Node value = Immediate(0);
value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdX), 0, 9);
value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdY), 16, 9);
value = BitfieldInsert(value, Operation(OperationCode::LocalInvocationIdZ), 26, 5);
return value;
}
case SystemVariable::TidX:
return Operation(OperationCode::LocalInvocationIdX);
case SystemVariable::TidY:
return Operation(OperationCode::LocalInvocationIdY);
case SystemVariable::TidZ:
return Operation(OperationCode::LocalInvocationIdZ);
case SystemVariable::CtaIdX:
return Operation(OperationCode::WorkGroupIdX);
case SystemVariable::CtaIdY:
return Operation(OperationCode::WorkGroupIdY);
case SystemVariable::CtaIdZ:
return Operation(OperationCode::WorkGroupIdZ);
default:
UNIMPLEMENTED_MSG("Unhandled system move: {}",
static_cast<u32>(instr.sys20.Value()));
return Immediate(0u);
}
}();
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::BRA: {
Node branch;
if (instr.bra.constant_buffer == 0) {
const u32 target = pc + instr.bra.GetBranchTarget();
branch = Operation(OperationCode::Branch, Immediate(target));
} else {
const u32 target = pc + 1;
const Node op_a = GetConstBuffer(instr.cbuf36.index, instr.cbuf36.GetOffset());
const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
PRECISE, op_a, Immediate(3));
const Node operand =
Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
branch = Operation(OperationCode::BranchIndirect, operand);
}
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
if (cc != Tegra::Shader::ConditionCode::T) {
bb.push_back(Conditional(GetConditionCode(cc), {branch}));
} else {
bb.push_back(branch);
}
break;
}
case OpCode::Id::BRX: {
Node operand;
if (instr.brx.constant_buffer != 0) {
const s32 target = pc + 1;
const Node index = GetRegister(instr.gpr8);
const Node op_a =
GetConstBufferIndirect(instr.cbuf36.index, instr.cbuf36.GetOffset() + 0, index);
const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
PRECISE, op_a, Immediate(3));
operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
} else {
const s32 target = pc + instr.brx.GetBranchExtend();
const Node op_a = GetRegister(instr.gpr8);
const Node convert = SignedOperation(OperationCode::IArithmeticShiftRight, true,
PRECISE, op_a, Immediate(3));
operand = Operation(OperationCode::IAdd, PRECISE, convert, Immediate(target));
}
const Node branch = Operation(OperationCode::BranchIndirect, operand);
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
if (cc != Tegra::Shader::ConditionCode::T) {
bb.push_back(Conditional(GetConditionCode(cc), {branch}));
} else {
bb.push_back(branch);
}
break;
}
case OpCode::Id::SSY: {
UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
"Constant buffer flow is not supported");
if (disable_flow_stack) {
break;
}
// The SSY opcode tells the GPU where to re-converge divergent execution paths with SYNC.
const u32 target = pc + instr.bra.GetBranchTarget();
bb.push_back(
Operation(OperationCode::PushFlowStack, MetaStackClass::Ssy, Immediate(target)));
break;
}
case OpCode::Id::PBK: {
UNIMPLEMENTED_IF_MSG(instr.bra.constant_buffer != 0,
"Constant buffer PBK is not supported");
if (disable_flow_stack) {
break;
}
// PBK pushes to a stack the address where BRK will jump to.
const u32 target = pc + instr.bra.GetBranchTarget();
bb.push_back(
Operation(OperationCode::PushFlowStack, MetaStackClass::Pbk, Immediate(target)));
break;
}
case OpCode::Id::SYNC: {
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "SYNC condition code used: {}",
static_cast<u32>(cc));
if (disable_flow_stack) {
break;
}
// The SYNC opcode jumps to the address previously set by the SSY opcode
bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Ssy));
break;
}
case OpCode::Id::BRK: {
const Tegra::Shader::ConditionCode cc = instr.flow_condition_code;
UNIMPLEMENTED_IF_MSG(cc != Tegra::Shader::ConditionCode::T, "BRK condition code used: {}",
static_cast<u32>(cc));
if (disable_flow_stack) {
break;
}
// The BRK opcode jumps to the address previously set by the PBK opcode
bb.push_back(Operation(OperationCode::PopFlowStack, MetaStackClass::Pbk));
break;
}
case OpCode::Id::IPA: {
const bool is_physical = instr.ipa.idx && instr.gpr8.Value() != 0xff;
const auto attribute = instr.attribute.fmt28;
const Tegra::Shader::IpaMode input_mode{instr.ipa.interp_mode.Value(),
instr.ipa.sample_mode.Value()};
Node value = is_physical ? GetPhysicalInputAttribute(instr.gpr8)
: GetInputAttribute(attribute.index, attribute.element);
const Tegra::Shader::Attribute::Index index = attribute.index.Value();
const bool is_generic = index >= Tegra::Shader::Attribute::Index::Attribute_0 &&
index <= Tegra::Shader::Attribute::Index::Attribute_31;
if (is_generic || is_physical) {
// TODO(Blinkhawk): There are cases where a perspective attribute use PASS.
// In theory by setting them as perspective, OpenGL does the perspective correction.
// A way must figured to reverse the last step of it.
if (input_mode.interpolation_mode == Tegra::Shader::IpaInterpMode::Multiply) {
value = Operation(OperationCode::FMul, PRECISE, value, GetRegister(instr.gpr20));
}
}
value = GetSaturatedFloat(value, instr.ipa.saturate);
SetRegister(bb, instr.gpr0, value);
break;
}
case OpCode::Id::OUT_R: {
UNIMPLEMENTED_IF_MSG(instr.gpr20.Value() != Register::ZeroIndex,
"Stream buffer is not supported");
if (instr.out.emit) {
// gpr0 is used to store the next address and gpr8 contains the address to emit.
// Hardware uses pointers here but we just ignore it
bb.push_back(Operation(OperationCode::EmitVertex));
SetRegister(bb, instr.gpr0, Immediate(0));
}
if (instr.out.cut) {
bb.push_back(Operation(OperationCode::EndPrimitive));
}
break;
}
case OpCode::Id::ISBERD: {
UNIMPLEMENTED_IF(instr.isberd.o != 0);
UNIMPLEMENTED_IF(instr.isberd.skew != 0);
UNIMPLEMENTED_IF(instr.isberd.shift != Tegra::Shader::IsberdShift::None);
UNIMPLEMENTED_IF(instr.isberd.mode != Tegra::Shader::IsberdMode::None);
LOG_WARNING(HW_GPU, "ISBERD instruction is incomplete");
SetRegister(bb, instr.gpr0, GetRegister(instr.gpr8));
break;
}
case OpCode::Id::MEMBAR: {
UNIMPLEMENTED_IF(instr.membar.type != Tegra::Shader::MembarType::GL);
UNIMPLEMENTED_IF(instr.membar.unknown != Tegra::Shader::MembarUnknown::Default);
bb.push_back(Operation(OperationCode::MemoryBarrierGL));
break;
}
case OpCode::Id::DEPBAR: {
LOG_DEBUG(HW_GPU, "DEPBAR instruction is stubbed");
break;
}
default:
UNIMPLEMENTED_MSG("Unhandled instruction: {}", opcode->get().GetName());
}
return pc;
}
} // namespace VideoCommon::Shader
|